光發射光譜大數據分析與機器學習輔助預測非晶矽與奈米晶氫化矽 (a–Si:H/nc–Si:H) 薄膜性質;Large scale data analysis and machine learning assisted prediction of a–Si:H to nc–Si:H transition based on Classiﬁers of OES，PCA and SVM

NCUIR > Engineering College > Graduate Institute of Mechanical Engineering > Electronic Thesis & Dissertation > Item 987654321/81623

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/81623

Title:	光發射光譜大數據分析與機器學習輔助預測非晶矽與奈米晶氫化矽 (a–Si:H/nc–Si:H) 薄膜性質;Large scale data analysis and machine learning assisted prediction of a–Si:H to nc–Si:H transition based on Classiﬁers of OES，PCA and SVM
Authors:	王淞禾;Wang, Song-Ho
Contributors:	機械工程學系
Keywords:	電漿輔助化學氣相沉積;三氯矽甲烷;氫氯化薄膜;光放射光譜;結晶相值;主成分分析;支持向量機;Plasma Enhanced Chemical Vapor Deposition (PECVD);Trichlorosilane (TCS, SiHCl3);Optical emission spectroscopy (OES);Value of crystalline phase (VCP);Hydrogen chloride silicon thin film;principal component analysis (PCA);Support vector machine(SVM)
Date:	2019-07-19
Issue Date:	2019-09-03 16:33:25 (UTC+8)
Publisher:	國立中央大學
Abstract:	電漿輔助化學氣相沉積法(Plasma Enhanced Chemical Vapor Deposition–PECVD)已被用於直接生產晶片同品質的矽薄膜元件，並利用三氯矽甲烷(Trichlorosilane–SiHCl3)為製程氣體源將製作成本降低。然而，氫氯化非晶矽薄膜為主要具有矽氫鍵，其結構相對於結晶矽鬆散，且矽氫鍵容易在照光後結斷裂形成缺陷而發生光衰效應，為提高薄膜穩定度，故研究目標為將薄膜由非晶結構轉至微晶或奈米晶結構，藉由調整 SiHCl3 與 H2 流量比例，直接沉積奈米晶矽結構薄膜，以提升矽薄膜光電元件之效率與穩定度。本研究採用PECVD以三氯矽甲烷氣體源在2項規劃的製程條件(射頻功率及氣體流量)和350℃的矽基板溫度下沉積氫氯化矽薄膜，並同時利用光發射光譜(Optical Emission Spectra–OES)監測紀錄沉積製程電漿的差異。接著使用傅里葉變換紅外光譜、拉曼光譜、穿透式電子顯微鏡、X–射線繞射分析和表面輪廓測量儀進行量測不同製程條件所沉積的氫氯化矽膜藉以分析薄膜的多種性質(包括有化學鍵結、結構(結晶狀態)和薄膜厚度(鍍率)等…)來確認其屬性–氫氯化非晶矽(a–Si：H/Cl)和氫氯化奈米晶矽(nc–Si：H/Cl)。所以為節省薄膜性質分析時所耗費的龐大成本與時間，透過機器學習技術可以解決複雜的問題，因而本研究結合主成分分析(PCA)與支持向量機(SVM)之機器學習技術，處理電漿製程之OES大規模數據，並建立自動化薄膜結晶結構分類與預測系統，可判斷不同製程所沉積出來的薄膜結構(氫氯化奈米矽晶、氫氯化非晶矽薄膜)。首先，處理電漿製程之OES大規模數據，應縮減複雜度高的OES全譜數據，選擇主要前驅物的自由基(SiCl2 ，SiCl3 ，Hα和Hβ)，並通過提出的PC1–DEV算法，建立結晶相值(Value of crystalline phase – VCP)，以表徵結晶結構的趨勢變化。本研究獲得高於0.06的VCP可以歸類為氫氯化奈米晶矽薄膜，在功率為250W流量為70sccm處有最大VCP值為0.23，其餘奈米晶矽薄膜的VCP平均值為0.11，另外低於0.06的VCP值將分類於非晶矽薄膜，然後將支持向量機法引入到對氫氯化矽薄膜處理的分類中，藉由使用三種不同的函數法線性函數核(Linear kernel)、多項式函數核(Polynomial kernel)和徑向基(高斯)函數核(Radial basis function kernel-RBF）演算法，並選出最佳訓練函數(徑向基函數核)，可建立出具有準確度為98%的高智慧判斷氫氯化矽晶薄膜結構表徵工具。 ;Plasma Enhanced Chemical Vapor Deposition (PECVD) has been used to improve the efficiency and stability of tantalum film optoelectronic components, including the nc–Si:H deposition film. In this paper, the nanocrystalline silicon thin films were deposited on Si substrate by PECVD from source gas of trichlorosilane (TCS, SiHCl3) at temperatures 350°C. The in–situ plasma monitoring and the resultant deposited film properties of Hydrogen chloride silicon thin film were characterized by Optical emission spectroscopy (OES), Fourier transfer infrared spectroscopy (FTIR), Raman spectroscopy (RS), X-ray Diffraction(XRD), Transmission electron microscope(TEM) and Alpha–Step profiler. In addition, principal component analysis (PCA) based on large scale OES dataset was performed and through the proposed PC1–DEV algorithm, the high–dimensional OES data of complexity should be selected and reduced to radicals of interest (SiCl2, SiCl3, Hα and Hβ). The value of crystalline phase (VCP) was established to differentially characterize the nanocrystalline phase as mean VCP of 0.11 and the control limits of 0.06, which can be used as the in–situ monitoring tool for crystalline phase characterization. And demonstrates use the large plasma data for PCA analysis connect with SVM algorithm method for the screening and grouping of nanocrystalline and amorphous OES spectra data, and make a decision with strong classifier performance. The support vector machine(SVM) method can classification of Hydrogen chloride silicon thin film, and using three different kernel function methods, include linear kernel, the polynomial kernel and the radial basis (Gaussian) function kernel. In this study the radial basis function kernel algorithm is the best training function used to be judgment and radial basis function kernel is selected to learn a high-smart judgment of the structure model of hydrogen chloride crystal film with an accuracy of 98%.
Appears in Collections:	[Graduate Institute of Mechanical Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	135	View/Open

社群 sharing

Loading...